Geophysical Journal InternationalGeophys. J. Int. (2018) 212, 942–954 doi: 10.1093/gji/ggx465Advance Access publication 2017 October 25GJI Seismology

Seismicity induced by longwall coal mining at the Thoresby Colliery,Nottinghamshire, U.K.

James P. Verdon,1 J-Michael Kendall,1 Antony Butcher,1 Richard Luckett2

and Brian J. Baptie2

1School of Earth Sciences, University of Bristol, Wills Memorial Building, Queen’s Road, Bristol BS8 1RJ, United Kingdom.E-mail: James.Verdon@bristol.ac.uk2British Geological Survey, The Lyell Centre, Research Avenue South, Edinburgh EH14 4AP, United Kingdom

Accepted 2017 October 24. Received 2017 October 20; in original form 2017 July 12

S U M M A R YThe United Kingdom has a long history of deep coal mining, and numerous cases of mining-induced seismicity have been recorded over the past 50 yr. In this study, we examine seismicityinduced by longwall mining at one of the United Kingdom’s last deep coal mines, the ThoresbyColliery, Nottinghamshire. After public reports of felt seismicity in late 2013 a local seismicmonitoring network was installed at this site, which provided monitoring from February toOctober 2014. This array recorded 305 seismic events, which form the basis of our analysis.

Event locations were found to closely track the position of the mining face within the DeepSoft Seam, with most events occurring up to 300 m ahead of the face position. This indicatesthat the seismicity is being directly induced by the mining, as opposed to being caused byactivation of pre-existing tectonic features by stress transfer. However, we do not observecorrelation between the rate of excavation and the rate of seismicity, and only a small portionof the overall deformation is being released as seismic energy.

Event magnitudes do not follow the expected Gutenberg–Richter distribution. Instead, theobserved magnitude distributions can be reproduced if a truncated power-law distribution isused to simulate the rupture areas. The best-fitting maximum rupture areas correspond to thedistances between the Deep Soft Seam and the seams that over- and underlie it, which haveboth previously been excavated. Our inference is that the presence of a rubble-filled void (orgoaf) where these seams have been removed is preventing the growth of larger rupture areas.

Source mechanism analysis reveals that most events consist of dip-slip motion along near-vertical planes that strike parallel to the orientation of the mining face. These mechanismsare consistent with the expected deformation that would occur as a longwall panel advances,with the under- and overburdens moving upwards and downwards respectively to fill the voidcreated by mining. This further reinforces our conclusion that the events are directly inducedby the mining process. Similar mechanisms have been observed during longwall mining atother sites.

Key words: Geomechanics; Earthquake source observations; Induced seismicity; Seismicanisotropy.

1 I N T RO D U C T I O N

Seismicity induced by coal mining has been a common occurrencein the United Kingdom (e.g. Redmayne 1988). Indeed, Wilson et al.(2015) estimated that between 20 and 30 per cent of all earthquakesrecorded in the United Kingdom between 1970 and 2012 wereinduced by coal mining. From the late 1980s onwards the rate of coalproduction has declined significantly, as has the rate of associatedearthquakes (Fig. 1).

Nevertheless, seismicity associated with deep coal mining stilloccurs in the United Kingdom. Between December 2013 and

January 2014, the United Kingdom’s national seismometer net-work detected a series of over 40 earthquakes near to the village ofNew Ollerton, Nottinghamshire. The largest of these events had amagnitude of ML = 1.7. Given the generally low levels of seismicityin the United Kingdom, the village was dubbed the ‘United King-dom’s Earthquake Capital’ (Turvill 2014). The area has a historyof seismic activity relating to coal mining (e.g. Bishop et al. 1993),and it was soon identified that the events were likely to be associatedwith longwall coal mining at the nearby Thoresby Colliery, whichat the time was one of the few remaining deep coal mining sites inthe United Kingdom.

942 C© The Author(s) 2017. Published by Oxford University Press on behalf of The Royal Astronomical Society.

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

Seismicity induced by longwall coal mining 943

Figure 1. Deep mined coal production in the United Kingdom by year (bars) and the number of induced earthquakes per year associated with coal mining(grey line), as categorised by Wilson et al. (2015). The drop in both production and induced seismicity in 1984 is associated with the U.K. miner’s strike.

In response to the felt earthquakes, a temporary local monitoringnetwork of surface seismometers was deployed between the 5thFebruary and the October 30th, 2014 by the British GeologicalSurvey (BGS). This network recorded a further 300 seismic events.The high quality of the data recorded by the local network permits adetailed study into the nature of seismicity and deformation inducedby the longwall mining process.

1.1 Longwall coal mining at Thoresby

The Thoresby Colliery opened in 1925. Over the history of the site,at least four different seams have been mined, including the HighHazels, Top Hard, Deep Soft and Parkgate Seams, in order fromshallowest to deepest: see Edwards (1967) for a stratigraphic sectionshowing the positions of these and other seams in the region. TheDeep Soft Seam was the last to be developed, with work beginningin 2010: this was the only seam being actively mined during thestudy period. The colliery closed entirely in mid-2015. This was foreconomic reasons related to the low price of coal, not because ofthe induced seismicity.

The Deep Soft Seam was mined using standard longwall methods:hydraulic jacks are used to support the roof while a shearing devicecuts coal from the face. As the face advances, the jacks are movedforward, allowing the roof to collapse into the cavity that is leftbehind. The collapsed, brecciated roof material filling this void isknown as goaf (e.g. Younger 2016). At Thoresby, each longwallpanel has dimensions of approximately 300 m width, between 1000and 3000 m length, and approximately 2.5 m height.

1.2 Seismicity associated with longwall coal mining

Seismicity has often been associated with the longwall mining pro-cess (e.g. Cook 1976; Gibowicz et al. 1990; Bishop et al. 1993; Stec2007; Bischoff et al. 2010; Sen et al. 2013). Seismic events associ-ated with coal mining have often been divided into two categories:‘mining-tectonic’ activity, produced by activation of pre-existingtectonic faults, and ‘mining-induced’ activity, directly associatedwith the mining excavations (e.g. Stec 2007).

Observed magnitudes have typically ranged from 0.5 < ML <

3.5. At some sites event magnitudes have followed the Gutenberg &

Richter (1944) distribution (e.g. Bishop et al. 1993; Kwiatek et al.2011), while in other cases bimodal or other frequency–magnitudedistributions have been observed (e.g. Stec 2007; Hudyma et al.2008; Bischoff et al. 2010). These non-Gutenberg–Richter distri-butions have been attributed to the presence of characteristic lengthscales (the dimensions of the mined panels, for example) that pro-vide a control on rupture dimensions and thereby event magnitudes.

Analysis of event focal spheres has revealed a variety of sourcemechanisms in different settings (e.g. Stec 2007; Bischoff et al.2010; Sen et al. 2013) including: non-double-couple events, indi-cating a volumetric component of deformation usually associatedwith the roof collapse process; double-couple events showing a di-rect relationship to mined panels, with vertical fault planes runningparallel to the mining face, on which dip-slip motion occurs; anddouble-couple events that correspond to regional fault orientationsand in situ tectonic stress conditions.

In this paper we follow the processes developed in the aforemen-tioned studies to characterize the seismicity induced by mining atthe Thorseby Colliery. We begin by locating events, comparing theevent locations to the propagation of the mining faces with time,and seismicity rates with the volume of coal extracted from themine. We investigate the source characteristics of the events, usingspectral analysis combined with event frequency–magnitude distri-butions to assess the length-scales of structures that have generatedthe observed events. We use shear-wave splitting analysis to imagein situ stress orientations at the site, and we calculate focal mecha-nisms for the events to establish the orientations of fault planes andslip directions generated by the mining process.

2 E V E N T D E T E C T I O N A N D L O C AT I O N

2.1 Monitoring array and event detection

The local surface network deployed to monitor seismicity at theThoresby Colliery comprised of 4 3-component Guralp 3ESPbroad-band seismometers (stations NOLA, NOLD, NOLE andNOLF) and 3 vertical-component Geotech Instruments S13J short-period seismometers (NOLB, NOLC and NOLG)). The stationpositions are shown in Fig. 2. Events were detected using theBGS’s in-house event detection algorithm, which is based on

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

944 J.P. Verdon et al.

Figure 2. Map of event hypocentres, with events coloured by occurrence date. Also shown are the positions of the monitoring network (triangles) and themining panels (brown rectangles). Panels DS-4 and DS-5 were active during the monitoring period, and the coloured bars running across these panels show theforward movement of the mining faces with time. The position of the cross-section A–B (Fig. 5) is marked by the dashed line.

Figure 3. Recorded waveforms for a larger event (ML = 1.3). The N (red), E (blue) and Z (green) components for each station are overlain. Stations NOLB,NOLC and NOLG are single (Z) component stations. The P- and S-wave picks are marked by the solid and dashed tick marks.

identification of peaks in running short-time/long-time averages(STA/LTA), as described by Allen (1982). A total of 305 eventswere identified during the deployment of the local monitoringnetwork.

P- and S-wave arrival times were repicked manually for everyevent (e.g. Fig. 3). For most event-station pairs the P-wave arrivalwas clear and unambiguous, and so could be accurately picked(83 per cent of station-event pairs where a pick could be manually

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

Seismicity induced by longwall coal mining 945

Table 1. 1D, layered, isotropic velocity model used to locate events. Modelis based on that used by Bishop et al. (1993).

Layer no. Depth to layer top (m) VP (ms−1) VS (ms−1)

1 0 1900 12802 60 2750 15403 135 3100 17404 275 3500 19705 1019 4200 23606 1351 5250 29207 2751 6000 3370

Figure 4. Histograms showing the lateral and depth uncertainties for thelocated events.

assigned). Stations NOLB, NOLC and NOLG were single, verti-cal component stations, so S-wave picks were not made for thesestations. For smaller events with lower signal-to-noise ratios, clearS-wave arrivals were sometimes difficult to identify, resulting in a

lower number of picks (74 per cent of station-event pairs where apick could be manually assigned).

The velocity model used to locate the events is taken from Bishopet al. (1993), and is listed in Table 1. The arrival time picks wereinverted for the best-fitting location that minimizes the least-squaresresidual between modelled and picked arrival times. The search forthe best-fitting location was performed using the NeighbourhoodAlgorithm (Sambridge 1999), and the modelled travel times werecalculated using an eikonal solver (Podvin & Lecomte 1991). Amap of event hypocentres is shown in Fig. 2, in which the miningpanels and the position of the mining face with time are also shown.

In Fig. 4 we show histograms of the event location uncertaintieslaterally and in depth. Note that these uncertainties pertain solely tothe residuals between picked and modelled arrival times, and do notaccount for velocity model uncertainties. The velocity model usedis based on limited site-specific data, relying mainly on regionalseismic refraction surveys (Bishop et al. 1993).

A brief sensitivity analysis suggested that velocity model uncer-tainties of up to 10 per cent may affect depth locations by as much as150 m, while lateral locations are relatively unaffected. This reflectsthe geometry of the array, which provides reasonable azimuthal cov-erage but with surface stations only, such that an uncertain velocitymodel will primarily affect the event depths.

Fig. 5 shows a cross-section of event depths relative to the coalseams. We note that, while it appears that the events are locatedbelow the seam depths, given the likely velocity model uncertainties;it is not possible to rule out that these events are actually located atthe same depths as the Deep Soft Seam being mined.

2.2 Event locations with respect to mining activities

The positions of the mining panels, and the progress of the miningface with time, have been provided by the UK Coal Authority intheir Mine Abandonment Plans (2017). The position of the miningface with respect to the events can be seen in Fig. 2. It is immediately

Figure 5. Events depths shown along cross-section A–B (see Fig. 2). The positions of the Top Hard, Deep Soft and Parkgate Seams are also marked. Note thatvelocity uncertainties mean that the event depths may not be particularly well constrained.

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

946 J.P. Verdon et al.

Figure 6. Histograms showing the lateral position of each event relative tothe mining face at the time of event occurrence, where a positive distancerepresents events occurring in advance of the face.

apparent that the event locations are tracking the position of the faceas it moves SE along panel DS-4, before switching to DS-5 and againtracking the mining front to the SE. The monitoring period ceaseswhen the events have propagated approximately half-way along thelength of panel DS-5.

We investigate the position of events in relation to the miningface in greater detail in Fig. 6, which shows a histogram of eventpositions relative to the mining face, along an axis parallel to themining panels. Most events are found to occur ahead of the face,with most events occurring within 300 m of the face. This closecorrelation between events and the mining face implies that theevents are being directly induced by mining activities, as opposedto the activation of pre-existing tectonic features, in which case wewould expect the events to align along an activated fault. As per thecategorisation described by Stec (2007), we characterise these asmining-induced events.

However, we also note small cluster of five events that is foundat greater depths (>2000 m), to the SW of the DS-4 panel. Four ofthese five events occurred within a single 7-hr period. Establishingthe causality of these events is more difficult. It is possible that theseevents have been have been triggered by the static transfer of stresschanges to greater depths, leading to fault activation. As per theStec (2007) categorisation, these may be mining-tectonic events.However, it is not possible to rule out that these deeper events mayin fact have a natural origin.

3 C O R R E L AT I O N B E T W E E NS E I S M I C I T Y A N D M I N I N G R AT E S ?

In Fig. 7 we show the volume of rock removed from the mine ona weekly basis (�V), the number of events per week (NE), and thecumulative seismic moment (

∑MO) released per week. The volume

of rock removed per week is estimated from the forward progressof the mining face, multiplied by its dimensions (width and height).To further investigate any correlation between the extracted volumeand seismicity, in Fig. 8 we cross-plot these parameters. From Fig. 8it is apparent that there is little immediate correlation between �Vand NE and

∑MO on a weekly basis.

McGarr (1976) posited a linear relationship between �V and∑MO:

�MO ≈ μ�V, (1)

where μ is the rock shear modulus. This relationship corresponds tothe situation whereby all of the deformation produced by the volumechange is released seismically. In reality, much of the deformationmay occur aseismically. As such, Hallo et al. (2014) proposed amodification to this relationship via a ‘seismic efficiency’ term,SEFF, which describes the portion of the overall deformation that isreleased as seismic energy:

�MO ≈ SEFFμ�V . (2)

In some of the most well-known cases of induced seismicity, valuesof SEFF have been close to 1 (e.g. McGarr 2014). However, thesecases represent outliers: during most industrial operations SEFF ismuch less that 1 (e.g. Hallo et al. 2014). The dashed lines in Fig. 8(b)show the relationship between �V,

∑MO and SEFF, assuming a

generic value of μ = 20 GPa. We note that the observed momentrelease rates correspond to values of SEFF between 0.01 and 0.00001,implying that most of the deformation induced by the mining isreleased aseismically. This is typical for many cases of seismicityinduced by a variety of industrial activities (e.g. Maxwell et al.2008; Hallo et al. 2014).

4 E V E N T M A G N I T U D E S A N DF R E Q U E N C Y– M A G N I T U D ED I S T R I B U T I O N S

4.1 Moment magnitude calculation

Local magnitudes for the Thoresby Colliery seismicity have beencomputed by Butcher et al. (2017), who found that the United King-dom’s existing local magnitude scale (Ottemoller & Sargeant 2013)

Figure 7. Weekly rock volume extracted (black lines) compared with (a) the weekly number of recorded events and (b) the weekly cumulative seismic momentreleased (grey lines).

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

Seismicity induced by longwall coal mining 947

Figure 8. Cross-plots examining potential correlation between weekly rock volume extracted and the weekly number of recorded events (a) and the weeklycumulative seismic moment released (b). In (b), the dashed lines show the expected relationship for given values of SEFF.

Figure 9. Example displacement spectrum used to estimate moment mag-nitudes. The solid line shows the observed spectrum, while the dashed lineshows the best-fitting Brune (1970) source model. The dot–dashed linesshow the fC and �O values for this model.

is not appropriate for use when sources and receivers are within afew kilometres of each other. This is because for nearby receivers,the ray path will be predominantly through the softer, more attenua-tive sedimentary cover, rather than the underlying crystalline crustalrocks, as will be the case for receivers that are more distant to theevent. Butcher et al. (2017) have developed an alternative localmagnitude scale based on the Thoresby events, which has been re-calibrated to ensure consistency between magnitude measurementsmade on nearby stations and those made using the UK’s permanentnational monitoring network, the nearest stations of which weresome distance from the Thoresby site.

However, our aim here is to investigate event magnitude distri-butions in order to understand the length scales of structures beingaffected by the mining process. This therefore requires the use ofmoment magnitudes, since seismic moment can be directly relatedto rupture dimensions. We compute moment magnitudes by fittinga Brune (1970) source model to the observed S-wave displacementamplitude spectra (Fig. 9), following the method described by Storket al. (2014). The seismic moment is determined from the amplitudeof the low-frequency plateau, �O.

Ideally, the measured corner frequency, fC, of the displacementspectra could be used to determine the rupture length. However,to robustly image the corner frequency, it must be significantlylower than the Nyquist frequency, fN of the recording system—Storket al. (2014) recommend that fN > 4fC to obtain robust estimates offC. The recording systems at Thoresby had sampling rates of 100Hz, so fN = 50 Hz.

We can use generic values for stress drop and rupture lengthsto establish the expected corner frequencies for events withMW < 1. Using the relationships between rupture dimensions, seis-mic moment and stress drop given by Kanamori & Brodsky (2004),assuming a stress drop of 5 MPa and a rupture velocity of 2000m s–1, the resulting corner frequency fC ≈ 30 Hz. Evidently, thefN > 4fC criteria is not expected to hold for this particular dataset.However, our observations of event magnitudes, because they arederived from the amplitude spectra at low frequencies, are robust:we therefore use these to make inferences about the length scalesof the structures that have generated the observed seismic events.

4.2 Frequency–magnitude distributions

The observed event magnitude distribution (EMD) is shown inFig. 10. We show the EMDs for the overall dataset, as well as indi-vidually for the clusters associated with the DS-4 and DS-5 panels.The overall dataset is not well described by the Gutenberg & Richter(1944) distribution log10 N (M) = a − bM , where N(M) is the cu-mulative number of events larger than a given magnitude M, and aand b are constants to be determined. Such a distribution would berepresented by a straight line in M versus log10(N) space. We notethat the apparent limit on the largest event size is not an artefactof a short measuring period: while the local array was removedin October 2014, the area continues to be monitored by the BGSNational Seismometer Array, which has an estimated detection ca-pability across the United Kingdom of magnitude >2. Larger eventsoccurring after the study period would therefore be detectable, butno such events have occurred.

However, fault length and/or earthquake magnitude distributionsthat are constrained at some upper limit, leading to a fall-off fromthe power-law (PL) relationship at large values, have been suggestedby a number of authors. At the largest scale, Richter (1958) arguesthat ‘a physical upper limit to the largest possible magnitude mustbe set by the strength of crustal rocks, in terms of the maximum

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

948 J.P. Verdon et al.

Figure 10. Observed frequency-magnitude distributions for the full eventpopulation (black), as well as for the DS-4 (light grey) and DS-5 (dark grey)clusters individually.

Table 2. Best-fitting power law and truncated power-law distributions foreach of the DS-4 and DS-5 clusters, and the resulting normalised misfits.

Dist. Type α C AMAX Misfit

DS-4 PL 0.47 1707 NA 5.46TPL 0.1 743 10075 1.23

DS-5 PL 0.74 6861 NA 3.05TPL 0.38 1536 3870 0.86

strain which they are competent to support without yielding’. Sim-ilarly, Pacheco et al. (1992) argue that the rupture dimensions ofvery large earthquakes are limited by the thickness of the Earth’sseismogenic zone (the portion of the crust that is capable of under-going brittle failure). For continental rift zones, Scholz & Contreras(1998) suggested that the maximum length of normal faults wouldbe limited by the flexural restoring stress and friction, and founda good match between their model and faults in the East AfricanRift and in Nevada. At a much smaller scale, Shapiro et al. (2013)have suggested these effects will also apply to induced seismicity,with the maximum fault size, and therefore earthquake magnitude,

determined by the dimensions of the volume stimulated by humanactivities.

To understand the observed EMDs at Thoresby, we consider thestatistical distributions of fault rupture areas that might producethem. Typically, rupture areas are assumed to follow a self-similar,PL distribution (e.g. Wesnousky et al. 1983; Bonnet et al. 2001). Ifstress drops are assumed to be roughly constant (e.g. Abercrombie1995) then this PL rupture area distribution will result in a PL dis-tribution of magnitudes, that is the Gutenberg–Richter distribution.

A cumulative PL distribution for rupture area will take the form:

N (L) = C A−α, (3)

where N(A) is the number of ruptures with area greater than lengthA, α is the PL exponent, and C is a constant. For a PL distribution,there is no upper limit to the maximum rupture area. Instead, ifan upper limit to the rupture area is imposed, for example by thegeometry of the mining panels, then a truncated power-law (TPL)distribution results (Burroughs & Tebbens 2001, 2002):

N (A) = C(

A−α − AMAX−α

), (4)

where AMAX is the maximum rupture area.To simulate event magnitudes based on rupture area, we use

Kanamori & Brodsky (2004):

MO = �σ A3/2, (5)

where �σ is the stress drop. As discussed above, the limitation ofa relatively low Nyquist frequency means that we cannot measurethe stress drop directly. Therefore, to estimate the PL and TPLparameters that best-fitting our observations, we initially assume ageneric and arbitrary stress drop of �σ = 5 MPa.

For each of the DS-4 and DS-5 event clusters, we perform a searchover the PL and TPL parameters, finding those that minimise theleast-squares misfit between observed and modelled EMDs. Theresulting EMDs are shown as the solid lines in Fig. 11, with the PLand TPL parameters, and the misfit for each of the models, listedin Table 2. The resulting rupture area distributions are shown inFig. 12.

Having established the best-fitting PL and TPL distributions witha fixed stress drop value, we then investigate the impact of a vari-able range of �σ . We do this in a stochastic manner, simulatingrupture area distributions based on the PL and TPL parameters,assigning stress drops randomly from a uniform distribution of

Figure 11. Fitting PL (black) and TPL (grey) rupture area distributions to the DS-4 (a) and DS-5 (b) EMDs. Observed EMDs are shown by black circles. Thesolid lines show the best-fitting models for a fixed �σ value, while the dashed lines show ±2 standard deviations when �σ is varied stochastically.

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

Seismicity induced by longwall coal mining 949

Figure 12. Best-fitting rupture area PL (black) and TPL (grey) distributions for the DS-4 (a) and DS-5 (b) clusters.

Figure 13. Diagrammatic section showing the spacing between the DeepSoft Seam, and the underlying Parkgate Seam, which has already beenmined out across the study area. Image taken from UK Coal Authority MineAbandonment Plans (2017).

0.1 < �σ < 20 MPa. We repeat this process over 100 iterations, andin Fig. 11 the dashed lines show the range encompassing ±2 stan-dard deviations around the resulting mean EMD. From Fig. 11 weobserve that both event populations are clearly better modelled by aTPL rupture area distribution, even when stochastic variation in �σ

is considered.Based on these results, it is worth examining whether the best-

fitting values for AMAX correspond to any length-scales associatedwith the mining activities. There are two length scales in play thatmight affect rupture dimensions: the width of the mining face (ap-proximately 300 m); and the separations between (1) the underlyingParkgate Seam, which is 35 m below the Deep Soft (Fig. 13), and(2) the overlying Top Hard Seam, which is approximately 110 mabove the Deep Soft. Both seams have already been mined through-out our study area. The voids left by the longwall mining of theseseams will be filled with goaf, the rubble and detritus created asthe roof collapses behind the mining face. It is difficult to envisagea mechanism by which ruptures could propagate through such arubble-filled void.

Assuming circular ruptures, areas of 10 075 and 3870 m2 corre-spond to rupture radii of 57 and 35 m. The larger dimension radiusis therefore roughly equivalent to a circular rupture extending fromthe Deep Soft to the Top Hard. Alternatively, assuming a rectangularrupture, the DS-4 AMAX value could correspond to a rupture withdimensions of approximately 35 × 300 m, equivalent to a ruptureextending from the Deep Soft to the Parkgate, across the lengthof the mined face. In reality, ruptures will not be rectangular norcircular. Nevertheless, the general agreement between the dimen-sions of the maximum rupture area and these distances leads usto suggest that the presence of the overlying and underlying TopHard and Parkgate seams is indeed limiting the rupture dimensions.Given the similarities between these dimensions, it is not possible todetermine whether one of these features in particular is controllingthe maximum rupture area. Indeed, it is likely that all three features:the width of the mining face; the distance to the underlying Park-gate Seam; and the distance to the overlying Top Hard Seam, are allplaying a role in limiting the maximum rupture dimensions.

5 S E I S M I C A N I S O T RO P Y A N DS H E A R - WAV E S P L I T T I N G

Shallow crustal anisotropy can be generated by several mechanisms,including: alignment of macroscopic fracture networks; the prefer-ential alignment of microcracks due to anisotropic stress field (inpractice, the microscopic and macroscopic effects often combine,as both larger-scale fracture networks and microcracks are pref-erentially opened or closed by the same stress field); and by thealignment of sedimentary bedding planes.

Shear-wave splitting (SWS), where the velocity of a shear-waveis dependent upon the direction of travel and the polarity of thewave, is an unambiguous indicator of seismic anisotropy, and hasbeen used previously to image stress changes induced by miningactivities (Wuestefeld et al. 2011). Shear waves that propagate near-vertically will not be sensitive to horizontally layered sedimentaryfabrics, which produce Vertically Transverse-Isotropy (VTI) sym-metry systems. Instead, in the absence of other major structuralfabrics, the fast shear wave polarisation orientation can be treatedas a proxy for the direction of maximum horizontal stress (e.g.Boness & Zoback 2006).

We perform SWS measurements on the Thoresby data. AccurateSWS measurements can only be obtained within the ‘S-wave win-dow’ (Crampin & Peacock 2008), because arrivals at an incidence

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

950 J.P. Verdon et al.

Figure 14. Example shear-wave splitting measurement using the method described by Teanby et al. (2004). In (a) we plot the N, E and Z components of therecorded waveforms, where P- and S-wave windows are highlighted by the shaded areas. In (b) we plot the radial and transverse components prior to andafter the splitting correction, where the aim of the correction is to minimise energy on the transverse component. In (c) we plot the waveform particle motionsbefore (solid lines) and after (dashed lines) correction. In (d) we plot the error surfaces of the correction method as a function of delay time and fast directionnormalised such that the 95 per cent confidence interval (highlighted in bold) is 1. In (e) we plot the best-fitting delay times and fast directions that result fromchoosing different S-wave window start and end times [as indicated by the light-grey shaded zone of (a)].

angle greater than ∼35◦ from vertical may be disturbed by S-to-Pconversions at the free surface. This constraint limits the avail-able data considerably, such that events within the S-wave windoware found only on station NOLA, and for only 28 of the recordedevents.

We perform the SWS measurement using the automated cluster-based approach described by Teanby et al. (2004). Where largerdatasets are studied, automated quality assessments such as thatdescribed by Wuestefeld et al. (2010) can be used, but in thiscase, given the small sample size, the quality of measurementswere assessed manually. Of the 28 arrivals within the S-wavewindow at NOLA, 9 provided good-quality, robust results ac-cording to the diagnostic criteria specified by Teanby et al.(2004). This is a typical rate-of-return for such studies given therelatively low magnitude (and therefore signal-to-noise) of theevents. An example of a robust SWS measurement is providedin Fig. 14.

In Fig. 15(a) we show the measured fast directions in the formof an angle histogram. A dominant fast direction striking NW–SEis clearly observed. The mean fast direction azimuth is 130◦. Notemporal variations in SWS fast directions or percentage anisotropywere observed. The mean delay time was 43 ms, and the meanpercentage S-wave anisotropy was 6.8 per cent.

In Fig. 15(b) we compare the measured fast S-wave orientationswith independent measurements for SHmax taken from the WorldStress Map database (Heidbach et al. 2008). These measurements,mainly from borehole breakouts and hydraulic fracturing tests, alsoindicate an approximate regional SHmax strike that is to the NW–SE.We conclude that the mean measured S-wave fast polarity of 130◦

can be used as a proxy for SHmax at this site.

6 S O U RC E M E C H A N I S M S

We compute event focal mechanisms by inverting the observed P-wave polarities and relative P wave, SH and SV wave amplitudesfor the best-fitting double-couple source mechanism. In doing so,we preclude the possibility of non-double-couple sources in ourinversion, as might be anticipated during mining-induced seismicity.We do this because the monitoring array consists of only 4 3-Cand 3 1-C stations, which limits our ability to robustly constrainnon-double-couple events. However, we note that the recoveredmechanisms do a reasonable job of fitting the observed polarities,non-double-couple sources do not appear to be necessary to matchthe majority of our observations.

Of the 305 events, a total of 65 had sufficient signal-to-noiseratios such that P-wave polarities could be robustly assigned, andproduced reliable and consistent source mechanisms. These strikes,dips and rakes for these events are plotted in Fig. 16. We note threemain clusters of event types, representative source mechanisms forwhich are also plotted.

The most common source mechanism type (numbered 1 inFig. 16) consists of events with strikes of approximately 50◦, highangles of dip and rakes of between 60◦ and 90◦. This source mech-anism orientation corresponds to near-vertical planes whose strikesmatch the strike of the mining face, on which dip-slip movementoccurs, with the side of the fault that is towards the mine movingdownwards.

A second, less populous source mechanism type (numbered 2 inFig. 16) shows similar strikes and dips, but with the opposite senseof movement such that the side of the fault towards the mining facemoves upwards. Similar event mechanisms—near-vertical failure

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

Seismicity induced by longwall coal mining 951

Figure 15. SWS and stress anisotropy. In (a) we plot an angle histogram of the measured SWS fast directions. In (b) we show regional measurements ofSHmax from the World Stress Map database (Heidbach et al. 2008): ‘+’ symbols represent borehole breakouts, ‘o’ symbols represent focal mechanisms, and‘�’ symbols represent hydraulic fracturing data. The Thoresby site is marked by the red square. Measurements are coloured by whether they represent a thrust,normal or strike-slip stress regime (if known).

Table 3. Principal stress orientations and Shape ratio (R) as inverted fromevent source mechanisms.

Stress Azimuth Plunge (down from horizontal) Shape ratio (R)

σ 1 144◦ 31◦ 0.17σ 2 52◦ 2◦σ 3 319◦ 59◦

planes striking parallel to the mining face with upward and down-ward dip-slip motion—were observed by Bischoff et al. (2010) forlongwall mines in the Ruhr Area, Germany, and we share their ge-omechanical interpretation for these events (Fig. 17). As the coal ismined, the surrounding rock mass will collapse to fill the void. Thiswill result in downward motion of the overlying rock (as per sourcemechanism type 1), and upward motion of the underlying rock (asper source mechanism type 2) along vertical planes that run parallelto the mining face.

A third type of source mechanism is also observed (numbered3 in Fig. 16), with thrust-type mechanisms occurring on steeplydipping planes that strike approximately north–south. It is possiblethat they result from the interaction between mining activities andpre-existing structures in the area, since the N–S orientation of theseplanes does not match the orientation of any feature in the mine.

Using the source mechanisms for all events, we use the STRESS-INVERSE iterative joint inversion algorithm described by Vavrycuk(2014) to estimate the orientations of principal stresses and the shaperatio, R (Gephart & Forsyth 1984):

R = σ1 − σ2

σ1 − σ3, (4)

where σ 1, σ 2 and σ 3 represent the maximum, intermediate andminimum principal stresses. The results of this inversion are listedin Table 3, and shown in Fig. 18. We note that the resulting maxi-mum horizontal stress is sub-horizontal, with an azimuth of 144◦.This is consistent, within error, with the maximum horizontal stressorientation estimated from SWS analysis. This implies that, whilethe orientations of the slip planes are consistent with the geometry

of the mining activities, the resulting deformation is also consistentwith the regional in situ stress conditions.

7 C O N C LU S I O N S

In this paper, we characterise the seismicity recorded during long-wall mining of the Deep Soft Seam at the Thoresby Colliery, Not-tinghamshire, UK. A local monitoring network was installed for 8months, recording 305 events, with the largest event having a localmagnitude of ML = 1.7. Event locations are found to track the ad-vance of the mining faces, with most events being located up to 300m ahead of the face.

We conclude that these events are ‘mining-induced’, that is theyare directly induced by the mining activity, as opposed to ‘mining-tectonic’ events, which are caused by static stress transfer producingactivation of pre-existing tectonic faults. However, comparison be-tween weekly mining rates and the rates of seismic activity do notshow strong correlation. Moreover, the amount of deformation re-leased in the form of seismic events is a small percentage of theoverall deformation produced by the mining activities (in otherwords, most of the deformation is released aseismically).

Event magnitudes do not follow the expected Gutenberg–Richterdistribution. Instead, we find that the observed magnitude distri-bution can be reproduced by assuming that rupture areas follow aTPL distribution, whereby there is a limit to the maximum size ofthe rupture area. The observed maximum rupture area could cor-respond to several controlling features around the seam, includingthe width of the mining face, and the distances to the underlyingParkgate and overlying Top Hard seams, which have already beenexcavated. Our inference is that the presence of these rubble-filledvoids where the excavated seams have been mined out creates alimit to the maximum rupture dimensions.

Event source mechanism analysis shows that most events com-prise dip-slip motion along near-vertical planes that strike parallelto the orientation of the mining face. This type of deformation isthe expected response to the longwall mining process, and has been

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

952 J.P. Verdon et al.

Figure 16. Source mechanisms (strike, dip and rake) for each event for which a reliable mechanism could be obtained. Three main clusters of mechanisms canbe identified, representative focal spheres for which are shown. These spheres are upper-hemisphere projections where the compressive quadrants are shadedblack.

Figure 17. Geomechanical interpretation of the observed source mechanisms. As the surrounding rocks move to fill the void created by mining, dip-slipmotion occurs on near-vertical slip planes oriented parallel to the mining face. Adapted from Bischoff et al. (2010).

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

Seismicity induced by longwall coal mining 953

Figure 18. Stress tensor inversion results using the STRESSINVERSE algorithm (Vavrycuk 2014). In (a) we show a lower hemisphere projection of the P(dark grey ©) and T (light grey ♦) axes for every event, with the overall estimate for the σ 1, σ 2 and σ 3 axes marked by a large ©, � and ♦, respectively. In(b) we show confidence limits for the principle stress axes, assuming ±15◦ error in source mechanism orientations.

observed at other longwall mining sites. The observed source mech-anisms are also consistent with the orientation of in situ regionalstresses as inferred from SWS analysis.

A C K N OW L E D G E M E N T S

JPV and JMK are funded by the BGS/University of Bristol StrategicPartnership in Applied Geophysics. This work was performed aspart of the Bristol University Microseismicity Project (BUMPS).

R E F E R E N C E S

Abercrombie, R.E., 1995. Earthquake source scaling relationships from −1to 5 ML using seismograms recorded at 2.5-km depth, J. geophys. Res.,100, 24 015–24 036.

Allen, R., 1982. Automatic phase pickers: their present use and futureprospects, Bull. seism. Soc. Am., 72, S225–S242.

Bischoff, M., Cete, A., Fritschen, R. & Meier, T., 2010. Coal mining inducedseismicity in the Ruhr area, Germany, Pure appl. Geophys., 167, 63–75.

Bishop, I., Styles, P. & Allen, M., 1993. Mining-induced seismicity in theNottinghamshire coalfield, Quart. J. Eng. Geol. Hydrogeol., 26, 253–279.

Boness, N.L. & Zoback, M.D., 2006. Mapping stress and structurally con-trolled crustal shear velocity anisotropy in California, Geology, 34, 825–828.

Bonnet, E., Bour, O., Odling, N.E., Davy, P., Main, I., Cowie, P. & Berkowitz,B., 2001. Scaling of fracture systems in geological media, Rev. Geophys.,39, 347–383.

Brune, J.N., 1970. Tectonic stress and the spectra of seismic shear wavesfrom earthquakes, J. geophys. Res., 75, 4997–5009.

Burroughs, S.M., 2002. The upper-truncated power law applied to earth-quake cumulative frequency-magnitude distributions: evidence for a time-independent scaling parameter, Bull. seism. Soc. Am., 92, 2983–2993.

Burroughs, S.M. & Tebbens, S.F., 2001. Upper-truncated power laws innatural systems, Pure appl. geophys., 158, 741–757.

Butcher, A., Luckett, R., Verdon, J.P., Kendall, J.-M., Baptie, B. & Wookey,J., 2017. Local magnitude discrepancies for near-event receivers: impli-cations for the U.K. traffic-light scheme, Bull. seism. Soc. Am., 107(2),532–541.

Cook, N.G.W., 1976. Seismicity associated with mining, Eng. Geol., 10,99–122.

Crampin, S. & Peacock, S., 2008. A review of the current understanding ofseismic shear-wave splitting in the Earth’s crust and common fallacies ininterpretation, Wave Motion, 45, 675–722.

Edwards, W.N., 1967. Geology of the country around Ollerton. Memoirs ofthe Geological Survey of Great Britain, Her Majesty’s Stationery Office.Available at: http://pubs.bgs.ac.uk/publications.html?pubID=B01568.

Gephart, J.W. & Forsyth, D.W., 1984. An improved method for determiningthe regional stress tensor using earthquake focal mechanism data: appli-cation to the San Fernando earthquake sequence, J. geophys. Res., 89,9305–9320.

Gibowicz, S.J., Harjes, H.-J. & Schafer, M., 1990. Source parameters ofseismic events at Heinrich Robert Mine, Ruhr Basin, Federal Republic ofGermany: evidence for nondouble-couple events, Bull. seism. Soc. Am.,80, 88–109.

Gutenberg, B. & Richter, C.F., 1944. Frequency of earthquakes in California,Bull. seism. Soc. Am., 34, 185–188.

Hallo, M., Oprsal, I., Eisner, L. & Ali, M.Y., 2014. Prediction of magnitudeof the largest potentially induced seismic event, J. Seismol., 18, 421–431.

Heidbach, O., Tingay, M., Barth, A., Reinecker, J., Kurfeß, D. & Muller, B.,2008. The World Stress Map Database Release 2008.

Hudyma, M., Potvin, Y. & Allison, D., 2008. Seismic monitoring of theNorthparkes Lift 2 block cave—Part 2. Production caving, J. S. Afr. Inst.Min. Metall., 108, 421–430.

Kanamori, H. & Brodsky, E.E., 2004. The physics of earthquakes, Rep.Prog. Phys., 67, 1429–1496.

Kwiatek, G., Plenkers, K. & Dresen, G., JAGUARS Research Group, 2011.Source parameters of picoseismicity recorded at Mponeng Deep GoldMine, South Africa: implications for scaling relations, Bull. seism. Soc.Am., 101, 2592–2608.

Maxwell, S.C., Shemeta, J., Campbell, E. & Quirk, D., 2008. Microseismicdeformation rate monitoring, in Proceedings of the SPE Annual TechnicalConference, Denver, SPE 116596.

McGarr, A., 1976. Seismic moments and volume changes, J. geophys. Res.,81, 1487–1494.

McGarr, A., 2014. Maximum magnitude earthquakes induced by fluid in-jection, J. geophys. Res., 119, 1008–1019.

Ottemoller, L. & Sargeant, S., 2013. A local magnitude scale ML for theUnited Kingdom, Bull. seism. Soc. Am., 103, 2884–2893.

Pacheco, J.F., Scholz, C.H. & Sykes, L.R., 1992. Changes in frequency-sizerelationship from small to large earthquakes, Nature, 355, 71–73.

Podvin, P. & Lecomte, I., 1991. Finite difference computation of traveltimesin very contrasted velocity models: a massively parallel approach and itsassociated tools, Geophys. J. Int., 105, 271–284.

Redmayne, D.W., 1988. Mining induced seismicity in UK coalfields identi-fied on the BGS national seismograph network, in Engineering Geologyof Underground Movements, eds Bell, F.G., Culshaw, M.G., Cripps, J.C.& Lovell, M.A., pp. 405–413, Geological Society Engineering GeologySpecial Publication no. 5.

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017

954 J.P. Verdon et al.

Richter, C.F., 1958. Elementary Seismology, Freeman and Co.Sambridge, M., 1999. Geophysical inversion with a neighbourhood

algorithm—I. Searching a parameter space, Geophys. J. Int., 138, 479–494.

Scholz, C.H. & Contreras, J.C., 1998. Mechanics of continental rift archi-tecture, Geology, 26, 967–970.

Sen, A.T., Cesca, S., Bischoff, M., Meier, T. & Dahm, T., 2013. Automatedfull moment tensor inversion of coal mining-induced seismicity, Geophys.J. Int., 195, 1267–1281.

Shapiro, S.A., Kruger, O.S. & Dinske, C., 2013. Probability of inducinggiven-magnitude earthquakes by perturbing finite volumes of rocks, J.geophys. Res., 118, 3557–3575.

Stec, K., 2007. Characteristics of seismic activity of the Upper Silesian CoalBasin in Poland, Geophys. J. Int., 168, 757–768.

Stork, A.L., Verdon, J.P. & Kendall, J.-M., 2014. The robustness of seis-mic moment and magnitudes estimated using spectral analysis, Geophys.Prospect., 62, 862–878.

Teanby, N.A., 2004. Automation of shear-wave splitting measurements usingcluster analysis, Bull. seism. Soc. Am., 94, 453–463.

Turvill, W., 2014. Welcome to Britain’s earthquake capital: Sleepy Not-tinghamshire town has been hit by 36 tremoers in just 50 days – andgeologists say mining is to blame, The Daily Mail, Available at: http://www.dailymail.co.uk/news/article-2548146/Welcome-Britains-EARTHQUAKE-capital-Sleepy-Nottinghamshire-town-hit-36-tremors-just-50-days-geologists-say-mining-blame.html, last accessed 21 January2017.

UK Coal Authority Mine Abandonment Plans, 2017. Mine Abandon-ment Plans are available upon application to the UK Coal Author-ity. Available at: https://www.gov.uk/guidance/coal-mining-records-data-deeds-and-documents.

Vavrycuk, V., 2014. Iterative joint inversion for stress and faultorientations from focal mechanisms, Geophys. J. Int., 199,69–77.

Wesnousky, S.G., Scholz, C.H., Shimazaki, K. & Matsuda, T., 1983. Earth-quake frequency distribution and the mechanics of faulting, J. geophys.Res., 88, 9331–9340.

Wilson, M.P., Davies, R.J., Foulger, G.R., Julian, B.R., Styles, P., Gluyas,J.G. & Almond, S., 2015. Anthropogenic earthquakes in the UK:a national baseline prior to shale exploitation, Mar. Petrol. Geol.,68, 1–17.

Wuestefeld, A., Al-Harrasi, O., Verdon, J.P., Wookey, J. & Kendall, J.M.,2010. A strategy for automated analysis of passive microseismic data toimage seismic anisotropy and fracture characteristics, Geophys. Prospect.,58, 755–773.

Wuestefeld, A., Kendall, J.M., Verdon, J.P. & Van As, A., 2011. Insitu monitoring of rock fracturing using shear wave splitting anal-ysis: an example from a mining setting, Geophys. J. Int., 187,848–860.

Younger, P.L., 2016. How can we be sure fracking will not pol-lute aquifers? Lessons from a major longwall coal mining analogue(Selby, Yorkshire, UK), Earth Environ. Sci. Trans. R. Soc. Edinburgh,106, 89–113.

Downloaded from https://academic.oup.com/gji/article-abstract/212/2/942/4564775by University Library useron 04 December 2017